.gitlab-ci.yml

include:
  - project: 'dl/devops/gitlab-ci-slurm'
    ref: master
    file: '/.gitlab-ci.yml'

stages:
  - build
  - test

variables:
  IMAGE_LATEST: "${CI_REGISTRY}/dl/hugectr/hugectr_inference_backend:${CI_COMMIT_BRANCH}.latest"

.cluster_test_job:
  extends: .dlcluster_job
  allow_failure: false
  stage: test
  dependencies: 
    - build
  # only:
  #   - merge_requests
  #   - web
  
### Stage: build
build:
  tags:
    - 1GPU
  stage: build
  script:
    - export IMAGE_VERSIONED="${CI_REGISTRY}/dl/hugectr/hugectr_inference_backend:${CI_COMMIT_BRANCH}.${CI_PIPELINE_ID}"
    - docker login -u ${CI_PRIVATE_USER} -p "${CI_PRIVATE_KEY}" "${CI_REGISTRY}"
    - export JOB_DOCKERFILE="Dockerfile.${CI_JOB_NAME%%--*}.${CI_PIPELINE_ID}" && echo ${JOB_DOCKERFILE}
    - cat ./inference.DockerFile > ${JOB_DOCKERFILE}
    - sed -i "s/https\:\/\/github.com\/NVIDIA\/HugeCTR.git/https:\/\/gitlab-ci-token:${CI_JOB_TOKEN}@gitlab-master.nvidia.com\/dl\/hugectr\/hugectr.git/g" ${JOB_DOCKERFILE}
    - sed -i "s/https:\/\/gitlab-master.nvidia.com\/dl\/hugectr\/hugectr_inference_backend.git/https:\/\/gitlab-ci-token:${CI_JOB_TOKEN}@gitlab-master.nvidia.com\/dl\/hugectr\/hugectr_inference_backend.git/g" ${JOB_DOCKERFILE}
    - sed -i "s/HUGECTR_BRANCH=main/HUGECTR_BRANCH=v3.0-integration/g" ${JOB_DOCKERFILE}
    - sed -i "s/INFERENCE_BRANCH=yingcan_integration/INFERENCE_BRANCH=${CI_COMMIT_BRANCH}/g" ${JOB_DOCKERFILE}
    - cat ${JOB_DOCKERFILE}
    - docker build --pull
        -t "${IMAGE_VERSIONED}"
        -f ${JOB_DOCKERFILE}
        ${PWD}
    - docker push "${IMAGE_VERSIONED}"
    - docker tag ${IMAGE_VERSIONED} ${IMAGE_LATEST}
    - docker push ${IMAGE_LATEST}
  only:
  #   - pushes
  #   - merge_requests
  #   - web

test:
  extends: .cluster_test_job
  script:
    - export IMAGE_LATEST_CORRECTE=$(echo $IMAGE_LATEST | sed 's/https:\/\///g')
    - srun -N 1 -p dgx1v,dgx1v16g,dgx1v32g bash -cx "
      echo $(pwd);
      cd ./test;
      docker login -u ${CI_PRIVATE_USER} -p "${CI_PRIVATE_KEY}" "${CI_REGISTRY}";
      bash ./triton_server.sh ${IMAGE_LATEST_CORRECTE} /mnt/nvdl/usr/aleliu/test/model_repository;
      docker run --rm --net=host -v /mnt/nvdl/usr/aleliu/test/inference_demo:/demo nvcr.io/nvidia/tritonserver:20.10-py3-clientsdk python3 /demo/hugectr_model1_request.py;
      docker run --rm --net=host -v /mnt/nvdl/usr/aleliu/test/inference_demo:/demo nvcr.io/nvidia/tritonserver:20.10-py3-clientsdk python3 /demo/identity_request.py;
      "